This page presents the work done during my PhD.
This work was accomplished into the CMI (Interactive conception and modelization) team of Ecole des Mines in Nantes (France).
Today, thanks to the developped algorithms and the constantly increasing computer power, computer generated images are widely available and can be generated in a short time. Their omnipresence, from virtual models of archeological sites to movies special effects, can hide that their modeling is actually time consuming. This modeling can be done by using classical tools based on geometrical primitives manipulation (shape modification, positioning ...). In that case, the user has to manage the whole modeling by himself: from objects modeling to their positioning. It makes him modify numerous parameters in order to keep the full scene coherent, when he wants to move or modify an object. As an example, if an object is smaller than initially modelised, objects lying on it have to be moved to reflect its size change. By using constraints based tools, the user can give constraints between objects properties (in most cases, numerical ones), the system insuring they are verified (sometimes approximatively when no exact solution exist). Examples of constraints can be something like "object A is higher than object B" or "the distance between objects B and C is less than d". In both types of tools, the user has to give more information than necessary because they do not have any semantics on objects and relations (in the constraints case, the user has to split high level symbolic relations into a set of constraints expressing a similar information). Yet, the description of an architectural scene by an architect to one of its co-workers is generally concise. This due to the fact that both parts (the architects) know the domain and so giving implicit information is useless since it is included into the semantics of domain objects and relations.
The goal of that work is to provide the user with an architectural reconstruction system allowing the historian and the architect to test the influence of hypothesis. Those hypothesis can be on a building construction (what happens if an object has the type X? for example) or in some cases to identify which one lead to the existing solution. Our work is based on the hypothesis that the use of the domain semantics can allow the user reconstruct scenes by specifying only useful information (not in the domain), but also to test different construction hypothesis. In that way, we have developped an approach based on the systematical use of a knowledge base in order to study its benefits and limits.
Didier Boucard, "Une méthode de reconstitution de scènes 3D fondée sur une base de connaissances. Application à l'architecture classique.", Thèse de doctorat, Université de Nantes, 26 novembre 2004. pdf (in french)
Didier Boucard, Christian Colin, "Vers un modeleur architectural cognitif : le projet MArCo", Revue internationale de CFAO et d'informatique graphique, volume 17, numéro 3-4, 2002.
Didier Boucard, "MArCo: A Knowledge-Based Architectural Reconstruction System for Hypothesis Testing", Europ'IA 10, Damascus, Syria, September 2005.
Didier Boucard, Stéphane Huot, Christian Colin, Gérard Hégron, Daniel Siret, "An image-based and knowledge-based system for efficient architectural and urban modeling", ACADIA 2002, Los Angeles, Octobre 2002.
Didier Boucard, Christian Colin, "MArCo, a knowledge-based architectural modeler and its application to the alignment problem", 3IA international conference on computer graphics and artificial intelligence, 14-15 May 2002, Limoges, France.
Didier Boucard, Stéphane Huot, Christian Colin, Gérard Hégron, Daniel Siret, "An image-based and knowledge-based system for efficient architectural and urban modeling", Colloque maquette virtuelle et patrimoine, ENSAM de Cluny, Mars 2003.
Didier Boucard, Christian Colin, "MArCo : vers un Modeleur Architectural Cognitif spécialisé en architecture du XVIIIe siècle", Colloque Modélisation multimodale appliquée à la reconstruction d'environnements architecturaux et urbains, 11-12 février 2002, Bordeaux.
Didier Boucard, Christian Colin, "Vers un modeleur architectural cognitif : le projet MArCo", Journées AFIG'2001, Limoges, 2001.
Gérard Hégron, Daniel Siret, Christian Colin, Pierre Macé,
Stéphane Huot, Eric Monin, Didier Boucard, "Une nouvelle approche
informatique de la modélisation architecturale : la restitution de
la place Ludovise de Louis", Colloque international Victor
Louis et son temps, Paris, Décembre 2000.